JP3278471B2 - Area division method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention identifies a character area and a non-character area (figure / table area) such as a figure area and a table area in an image of a general document in which figures and tables are mixed. The present invention relates to an area dividing method for classifying and performing necessary area division.

[0002]

2. Description of the Related Art When filing a document image on a storage medium such as an optical disk, communicating by facsimile, or outputting to a printer, the document image is divided into a character area and a figure / table area. In some cases, it is desired to perform area processing, and to perform figure processing and table processing on a figure / table area. When encoding characters on a document using an optical character recognition device, it is necessary to separate the same character region from the figure / table region.

Conventionally, as a technique relating to such an area extraction, a circumscribed rectangle of a black connected component is extracted from a document image, and the size of the extracted rectangle is compared with a predetermined threshold value. Is known (Japanese Patent Application Laid-Open No. 55-162177).

[0004]

However, since a predetermined threshold value is used for determining the type of the circumscribed rectangle of the black connected component, it is not possible to flexibly cope with input of various document images having different character sizes.

When a document to be divided into regions is a book or magazine page, it is rare to cut out one page at a time and set it on the scanner. Open the target page of the book or magazine and set it on the scanner as a two-page spread document. Often read. In the case of an image obtained by reading such a two-page spread original, a darkened or black area is likely to occur at the center folded portion of the original, that is, at the boundary between pages. In addition, the scanner holding plate (for reflecting the light from the scanner light source to make the background of the image white) cannot be closed, or it rises, resulting in a black area around the original. Cheap. In the case of performing area division of such a two-page spread original image, a black area observed in a bent portion or a peripheral part should be appropriately processed as a noise area.
This noise processing was not considered.

In a document image, a character area (column) may be close to a ruled line. However, conventionally, no consideration has been given to a ruled line as a separator for such a character area. In some cases, an adjacent character area delimited by, may be integrated into one character area, or a character area in a table area or a character area in a graph area may be integrated with an external character area adjacent thereto.

An object of the present invention is to provide an area dividing method for separating a character area with higher accuracy. More specifically, one object of the present invention is to provide an area dividing method for accurately separating a character area and other areas from a document having various character sizes and the like and performing necessary area division. It is to provide. Another purpose is to eliminate the influence of noise areas that are likely to occur in the center-folded area and peripheral area of a document image read from a spread original such as a book or a thick original, and to perform an accurate area division. It is to provide a division method. It is another object of the present invention to provide an area dividing method capable of accurately separating a character area delimited by ruled lines, a character area inside a table area or a graph area.

[0008]

According to the first to third aspects of the present invention, a standard character size is determined and extracted from a histogram of the height of a rectangle circumscribing a black connected component basically extracted from a document image. Rectangles are classified into character rectangles and other rectangles based on the size relationship with the standard character size,
Then, a character area is generated by integrating only the character rectangles.

According to the fourth aspect of the present invention, a rectangle other than a character in contact with a boundary of a processing target area of a document image is removed as noise.

According to the fifth aspect of the present invention, in addition to extracting a character rectangle from a document image and extracting a vertical ruled line and integrating rectangles from a given rectangle, the horizontal distance from the rectangle of interest is When a vertical ruled line that is smaller than the threshold of the distance between rectangles, which is a horizontal integration condition, and that vertically overlaps with the rectangle of interest is encountered, regarding the region of interest,
The threshold value of the distance between rectangles, which is a horizontal integration condition, is changed to a value corresponding to the horizontal distance between the target rectangle and the vertical ruled line.

According to the sixth aspect of the present invention, lines generated by merging rectangles of characters are merged. When lines are merged from a line of interest, the line to be merged is vertical at the time of generation. When the threshold value of the distance between rectangles, which is a horizontal integration condition, is changed due to an encounter with a ruled line, integration is prohibited when the horizontal end point coordinates of the line of interest are larger than the horizontal end point coordinates of the line.

According to the seventh aspect of the present invention, the character area generated by the line integration is reintegrated. When the character area is integrated from a certain noted character area, the character area within the character area to be integrated is re-integrated. If the line includes a line in which the threshold value of the distance between rectangles, which is a horizontal integration condition, is changed due to encountering a vertical ruled line at the time of generation, the horizontal end coordinates of the character area to be integrated are calculated from the horizontal end coordinates. When the coordinates of the end point in the horizontal direction of the character area of interest are large, integration is prohibited.

According to the eighth aspect of the present invention, in addition to extracting a character rectangle from a document image, extracting a vertical ruled line, and performing line integration from a line of interest for a line obtained by integrating character rectangles, When the vertical distance to the line of interest is smaller than the threshold of the line distance which is a vertical integration condition, and a horizontal ruled line having a horizontal overlap with the line of interest is encountered,
For the target line, the threshold value of the distance between lines, which is a vertical integration condition, is changed to a value corresponding to the vertical distance between the target line and the horizontal ruled line.

According to the ninth aspect of the present invention, the character area generated by the line integration is re-integrated from the focused character area. At this time, the line within the character area to be integrated is included in the line. If a line in which the threshold value of the line spacing as a vertical integration condition is changed due to encounter with a horizontal ruled line at the time of generation is included, the vertical direction of the target character region is calculated from the vertical end coordinates of the character region to be integrated. When the end point coordinates are large, integration is prohibited.

According to the tenth or eleventh aspect , in addition to extracting a character rectangle from a document image, a large figure / table rectangle other than a character is extracted. According to the tenth aspect of the present invention, the figure / table rectangular rectangle is scanned in the horizontal direction to extract a rectangle H circumscribing a black connected component consisting of only black runs longer than the threshold, and extracting a rectangle H other than each character. Among the rectangles H extracted from the rectangles, those that are located at the top or bottom within a predetermined distance from the upper side or lower side of the rectangle other than the character and satisfy the shape conditions as horizontal ruled lines are imaginary horizontal separators. And treats this imaginary horizontal separator as a horizontal separator that vertically separates the character area, and integrates the character rectangles. According to the eleventh aspect , the figure / table rectangle is scanned in the vertical direction to extract the rectangle V circumscribing the black connected component consisting of only the black runs longer than the threshold, and is extracted from the rectangles other than the respective characters. Rectangle V
Among them, those which are located at the leftmost or rightmost within a predetermined distance range from the left side or right side of the rectangle other than the character and satisfy the shape conditions as vertical ruled lines are extracted as imaginary vertical separators, and this imaginary vertical separator is extracted. The separator is treated as a vertical separator that separates the character area to the left and right, and character rectangles are integrated.

[0016]

According to the first to third aspects of the present invention, character rectangles and other rectangles (chart rectangles) can be accurately classified with respect to images of documents having various character sizes. The accuracy of extracting a character area that is a set of character rectangles is improved.

According to the fourth aspect of the present invention, even for a document image read from a book or a thick original, the influence of a black area generated at a bent portion or a peripheral portion is eliminated, and a character rectangle, a line, or a character is read. The region can be accurately extracted.

According to the fifth to seventh aspects of the present invention, erroneous integration of character areas (columns) separated by the vertical separator can be prevented, and each character area can be correctly separated.

According to the eighth or ninth aspect of the present invention, erroneous integration of character regions separated by the horizontal separator can be prevented, and each character region can be correctly separated.

According to the tenth or eleventh aspect of the present invention, it is possible to prevent erroneous integration between a table or graph region and a character region approaching the region, and to correctly extract a character region.

Hereinafter, various embodiments will be described with reference to the drawings in order to clarify the features and advantages of the present invention.

[0022]

FIG. 1 is a block diagram for explaining Embodiments 1 to 6 to be described later.

Reference numeral 1 denotes a document image memory for storing a document image read by a scanner. The area to be processed for the document image may be the entire document image or an area specified by displaying the document image on a display screen and using a mouse or the like.

A rectangle extracting unit 5 extracts a rectangle circumscribing the black connected component of the input document image. However, an image compression means may be provided at a stage preceding the rectangle extracting unit 5 to perform rectangle extraction on an image obtained by compressing an input document image (for example, an image obtained by compressing 8 × 8 pixels into one pixel). Reference numeral 2 denotes a rectangular memory for storing information on the extracted rectangle.

Reference numeral 6 denotes a histogram generation unit for generating a histogram of a rectangular height, reference numeral 3 denotes a histogram memory for storing information of the histogram, and reference numeral 7 denotes a standard character size for determining a standard character size from the histogram of the rectangular height. The determining unit 4 is a standard character size holding memory for storing information on the determined standard character size.

Reference numeral 8 denotes a rectangle classifying unit for classifying the extracted rectangles into character rectangles and other rectangles (rectangles in the figures). Reference numeral 9 denotes a line extracting unit that integrates character rectangles and extracts lines of a character string, and 12 denotes a line memory for storing information on the extracted lines. Reference numeral 10 denotes a character area generation unit that integrates lines to generate a character area that is a set of lines, and 13 denotes an area memory for storing information on the generated character areas. A control unit 11 controls the operation of each unit.

Each processing unit and control unit 11 may be realized by hardware, may be realized by software on a computer system, or may be realized by a combination of hardware and software. Which of the implementation forms should be selected in consideration of the required processing speed and the like.

FIG. 2 is an explanatory diagram of the processing outline. When the original image shown in (a) is input, rectangle data shown in (b) is obtained by the rectangle extraction processing. (D) shows the enlarged rectangular data. By performing rectangle integration and line integration on the rectangular data, line / region data shown in (c) is obtained. Hereinafter, the contents of such processing will be described for each embodiment.

Embodiment 1 FIG. 3 shows a processing flow from rectangle extraction to determination of a standard character size. Step 100 is a process of the rectangle extracting unit 5 for extracting a black link on the input document image and extracting a rectangle circumscribing the black connection, for example, using the coordinates of the upper left corner and the lower right corner of the rectangle and the rectangle number as rectangle information. It is stored in the rectangular memory 2.

Step 105 is a process of the histogram generation unit 6, which is based on the rectangle information in the rectangle memory 2 and which is the height of the rectangle (the size of the rectangle in the vertical direction with respect to the line. Generate a histogram (which is the size in the horizontal direction). FIG. 4 shows an example of this histogram.

Steps 110 to 115 are processing of the standard character size determination unit 7. In the case of this embodiment, the total frequency (the number of rectangles) of the histogram of the height of the rectangle is T, and t = T /
16 is calculated (step 110). Next, based on the histogram of the heights of the rectangles, the maximum rectangle height among the rectangle heights whose frequency exceeds t is calculated as the standard character size siz of the document.
e (steps 111 to 115). Therefore, when the histogram shown in FIG.
It is determined that ze = 4, and when the histogram shown in (b) is obtained, it is determined that size = 6.

Next, rectangle classification is performed by the rectangle classification unit 8 using the standard character size size as a reference. For example, a rectangle whose height and width (size in the line direction) do not exceed 5 times the standard character size size is classified as a character rectangle, and a rectangle larger than that is a non-character rectangle (figure / table rectangle). ).

The result of this classification is the output of this embodiment. That is, the document image is divided into a character area and a figure / table area using the rectangle of the black connected component as a unit. In FIG. 2, the rectangular data in FIG. 2B is the processing result of the present embodiment.

Embodiment 2 The processing up to the rectangle classification by the rectangle classification unit 8 is the same as that of the first embodiment.
However, the following row extraction is performed by the row extracting unit 9.

The distance (line direction) between two character rectangles is obtained,
If the distance is equal to or smaller than the standard character size, both character rectangles are integrated. In the case of a horizontally written document, for example, the horizontal distance sp1 of the character rectangles 151 and 152 shown in FIG.
If 1 ≦ size, they are integrated. In the case of a vertically written document, similar integration is performed according to the vertical distance. Such integration processing is performed on all character rectangles, and a circumscribed rectangle of the integrated character rectangle group is cut out as a character string line.
In FIG. 6, reference numeral 153 denotes a cut-out line, and a rectangle inside the line is a character rectangle.

As described above, in this embodiment, the character rectangles are integrated, and the character area of the document image is extracted in units of lines. In FIG. 2, the data at the stage of extracting a line by integrating the character rectangle with the rectangular data shown in FIG.
It is a processing result.

Embodiment 3 The method of integrating character rectangles after rectangle classification and extracting lines is the same as in the second embodiment, but the method of integrating character rectangles is different.

That is, in the case of a horizontally written document, for example, FIG.
The overlap amount sp2 of the character rectangles 161, 162 is measured,
Assuming that the smaller value of the rectangular heights h1 and h2 is h, sp
If 2 ≧ h / 2, the character rectangles 161 and 162 are integrated.
The circumscribed rectangle of the character rectangle group integrated in this way is extracted as a line.

In the case of a vertically written document, the same determination is made using the overlap amount between the character rectangles in the vertical direction and the height (the size in the horizontal direction) of the character rectangles, and the lines are extracted by integrating them.

Embodiment 4 A line extraction unit 9 performs line extraction by integrating character rectangles.
At this time, when both the integration condition of the character rectangle in the second embodiment and the integration condition of the character rectangle in the third embodiment are satisfied, the integration of the character rectangle is performed.

Fifth Embodiment Processing until a row is extracted by the row extracting unit 9 is the same as that of the second, third or fourth embodiment.

Next, the character area generating unit 10 generates a character area (column) by integrating lines as described below. First, size is set to p as an initial value, and when the distance p2 from the row to be integrated is equal to or less than p, the row is integrated, but when p2 exceeds p, the integration is not performed.

When the rows are integrated, a value p that is 1.5 times the distance p1 (the distance in the direction perpendicular to the rows) between the integrated rows.
(If p1 is a certain value or less, for example, 1 or less, p
Is set to a fixed value, for example, 3). When the distance p2 to the next row to be integrated is p or less, the rows are integrated, and when the distance exceeds p, the integration is not performed.

However, even if p2 ≦ p, the height h1 of the last row to be integrated and the height h2 of the row to be integrated next
Is not integrated when the ratio h2 / h1 is equal to or greater than a certain constant (for example, 5/4).

FIG. 8 is an explanatory diagram of the first integration condition. However, this is the case for horizontal writing documents. Rows 171, 172
And then for p2 ≦ p for row 173,
Integrate this. As a result, a character area 174 is obtained.

FIG. 9 is an explanatory diagram of the second integration condition. When the rows 181 and 182 are integrated and the row 183 is integrated next, h2 / h1 ≧ 5/4 is not integrated.
As a result, a character area 186 of rows 181 and 182 is obtained. In this example, in addition to the character region 186, a character region 187 of only the line 183 and a character region 188 obtained by integrating the lines 184 and 185 are generated.

As described above, according to this embodiment, it is possible to extract a character area in which lines are collected from a document image.
In FIG. 2, the row / area data of (c) is obtained.

Embodiment 6 The character area generation unit 10 generates a character area by line integration similar to that of the fifth embodiment, but further integrates the following character areas in order to complete the integration. Do.

The width of the two character areas to be integrated is
Both are more than 5 times the standard character size size,
The integration conditions are different when at least one of the widths is smaller than five times the standard character size size.

FIG. 10 is an explanatory diagram of the former case, that is, the case of integrating large areas. 191 and 192 are character areas generated by the same integration processing as in the fifth embodiment. The overlap amount of the character areas 191 and 192 in the row direction is represented by s.
p1, width of the character area 191 is w, character areas 191, 19
2 is sp2, sp1 ≧ 4 / 5w, and s
When p2 <size, the character regions 191 and 192 are integrated into one character region 193.

FIG. 11 is an explanatory diagram of the latter case of integrating small areas. 195 and 196 are character areas to be integrated. The distance in the line direction of the character regions 195 and 196 is sp1, the height of the smaller character region 196 is h, the amount of overlap in the vertical direction with respect to the lines of the character regions 195 and 196 is h / 2 or more, and sp1 ≦ In the case of 2 × size, they are integrated into one character area 197.

Embodiment 7 FIG. 12 is a block diagram for explaining the present embodiment .
FIG. 1 shows that the noise rectangle removing unit 201 is added.
And different.

As described above, when the original of the document image input to the document image memory 1 is a book or a thick original, a black area or a dark area is formed in the middle bent portion or the peripheral portion of the original. In the present embodiment, a figure / table rectangle (a rectangle other than a character rectangle) corresponding to such a black area is removed as noise, and a means provided for that purpose is a noise rectangle removal unit 201.

FIG. 13 is a processing flowchart of the noise rectangle removing unit 201. Referring to the rectangle memory 2, the rectangle classification unit 8
The information of the classified rectangle is read out to check whether it is a figure / table rectangle (a rectangle other than a character rectangle) (step 21).
0), if it is a figure / table rectangle, it is checked whether it is in contact with any boundary side of the processing target area (step 21).
5). Here, the processing target region is the entire document document or a designated region. If it touches the boundary side, the figure / table rectangle is removed as a noise rectangle. The removed rectangle is excluded from the target rectangle for the next row extraction. example
For example, as shown in FIG.
Since it is in contact with the area 230, it is excluded.

FIG. 15 is a block diagram for explaining Embodiments 8 to 12 to be described later. Reference numeral 301 denotes a document image memory for storing input document image data; 306, a rectangle extraction unit that extracts a circumscribed rectangle of a black connected component of the input document image; and 302, a rectangle memory that stores information on the extracted rectangle. is there.

The process of extracting a rectangle may be the same as in the first to seventh embodiments. For example, a circumscribed rectangle of a connected black run may be extracted while scanning a document image, and the rectangles within a certain distance may be integrated. Performed by Note that, for example, 8 × 8 pixels of the input document image
A means for compressing (reducing) pixels may be provided, and a rectangle may be extracted from the compressed image.

Reference numeral 307 denotes a rectangle classifying unit for classifying rectangles into character rectangles and other figure / table rectangles. Processing of the rectangular classification, similarly as in Example 1 to 7, is performed by comparing the rectangle width and height as the threshold value, the threshold value at that time is given in advance, unlike Embodiments 1 to 7 . However, similarly to the first to seventh embodiments, for example, a standard character size may be obtained from a histogram of a rectangular height and a threshold value may be adaptively calculated by multiplying the standard character size by an appropriate coefficient.

Reference numeral 308 denotes a vertical ruled line identification unit for identifying a vertical ruled line (vertical separator). This vertical ruled line identification
For example, the following process is performed. The width (horizontal size) of the figure / table rectangle (large rectangle other than the character rectangle) classified by the processing of the rectangle classification unit 307
Is selected as a candidate for a vertical ruled line having a height smaller than a predetermined threshold and a height (vertical size) larger than the predetermined threshold. Next, by scanning the document image in the range of the vertical ruled line candidate rectangle in the vertical direction, only black runs longer than a predetermined threshold are extracted, and a circumscribed rectangle of a connected component of only the long black runs is extracted. Then, between the length H1 and width W1 of the circumscribed rectangle of this long black run connected component and the height H and width W of the original rectangle, H1 / H> threshold (for example, 0.8) and W1 / If W> threshold (for example, 0.8), it is determined that the circumscribed rectangle of the long black run connected component is a vertical ruled line.

It should be noted that a threshold value for determining a vertical ruled line rectangle,
The threshold value of the length of the black run and the length of the black run may be adaptively determined similarly to the threshold value of the rectangle classification. The specification and drawings of Japanese Patent Application No. 4-160866 show such a ruled line extraction method in more detail.

Reference numeral 309 denotes a line extracting unit that generates a line by integrating character rectangles, and reference numeral 303 denotes an integrated threshold memory that stores a threshold value for judging the integration of character rectangles at the time of line generation. 3
Reference numeral 04 denotes a row memory for storing information on the generated rows.
A character area generation unit 310 integrates the generated lines to generate a character area. An area memory 305 stores information on the generated character area. The process of line unification is
As in the first to seventh embodiments, the integration is determined by comparing the distance between rows to be integrated with the integration threshold, and integration is performed by a method of integrating possible rows. At this time, the presence / absence and relative positional relationship of the vertical ruled line extracted by the vertical ruled line identification unit 308 are referred to. Reference numeral 311 denotes a character region correction unit that corrects a character region by re-integrating a region that has not been completely integrated into a character region generated by the line integration of the character region generation unit 310. Reference numeral 312 denotes a control unit that controls the above-described units.

Hereinafter, Examples 8 to 12 will be described individually. However, it is assumed that the lines on the document are horizontal.

Embodiment 8 Rectangle extraction processing in rectangle extraction section 306, rectangle classification section 307
, A vertical ruled line is extracted by the vertical ruled line identifying unit 308, and then a line is generated by a line extracting unit 309.

FIG. 16 shows a schematic flow of the row generation processing.
The basic processing of line generation is as follows. For a rectangle determined to be a character rectangle, a horizontal distance (sp1) and a vertical distance (sp2) between a given rectangle and another rectangle, a standard character size, etc. From the horizontal integration threshold (Th1) and the vertical integration threshold (T
h2), the adjacent rectangles are integrated. As shown in FIG. 16, the horizontal integration threshold is changed according to the relationship with the vertical ruled line.

First, a new character rectangle A is selected (step 360), and the horizontal integration threshold is set to Th1 (step 3).
62). A rectangle (character rectangle or vertical ruled line rectangle) is searched rightward from the focused character rectangle A in a range where the horizontal distance is within Th1 and the vertical distance is within the vertical integration threshold Th2,
It is determined whether the found rectangle is a vertical ruled line rectangle (step 364). If the rectangle is a character rectangle, they are integrated (step 366), and a search for the next rectangle is performed.

Since the purpose is to integrate character rectangles, if the searched rectangle is a vertical ruled line rectangle, it is not naturally integrated with the vertical ruled line rectangle. Then, it is checked whether or not the rectangle A of interest and the vertical ruled line rectangle overlap in the vertical direction (step 368). If there is an overlap, the distance Th1 'between the rectangle A and the center line of the vertical ruled line rectangle is set horizontally. Reset to the integration threshold (step 37)
0), search for the next rectangle.

When the process is repeated for the rectangle A until the next rectangle is not found within the range below the horizontal integration threshold (Th1 or Th1 ') and below the vertical integration threshold Th2 (step 372), another attention rectangle from step 360 is obtained. The processing is started for.

A specific example of a case where a vertical ruled line is encountered during the integration of a character rectangle will be described with reference to FIG. In FIG. 17A, first, a rectangle B is integrated with a rectangle A. Next, a vertical ruled line rectangle 3 within a horizontal distance of Th1 or less from rectangle A
80 is found. Since the vertical ruled line rectangle 380 overlaps with the rectangle A, the horizontal integration threshold is changed from Th1 to Th1 ′. As a result, if the horizontal integration threshold is Th1, the rectangle C that was to be searched and integrated next. Are excluded from the integration target with the rectangle A. Therefore, the row 381 composed of the rectangles A and B and the rectangle C separated by the vertical ruled line (vertical separator) are correctly separated.

In FIG. 17B, a rectangle B is integrated with a rectangle A, and then a vertical ruled line rectangle 382 is searched. However, since the vertical ruled line rectangle 382 does not overlap with the rectangle A, the horizontal integration threshold remains at Th1. Therefore, the next rectangle C is searched and integrated with the rectangle A, and a row 383 including rectangles A, B, and C is generated. That is, this vertical ruled line rectangle 382 is at least a rectangle B
Since it is not a vertical separator between the character and the rectangle C, the character rectangle is integrated ignoring its existence.

Ninth Embodiment As in the eighth embodiment , the horizontal integration threshold value is changed when a vertical ruled line rectangle overlapping the character rectangle of interest exists in the line generation processing. Are different. Except for this, it is the same as the eighth embodiment .

FIG. 18 is an explanatory diagram of a method of determining a change value of the horizontal integration threshold. An image is horizontally scanned in a vertical range (Ya1 to Ya2) of the rectangle A and a scan range 385 determined by a horizontal range (xr1 to Xr2) of the vertical ruled rectangle 384. The existence position (Xr3) of the vertical ruled line is obtained, and the distance (Xr3) is obtained.
r3-Xa2) is set as the changed horizontal integration threshold. Alternatively, the average value of the positions of the vertical ruled lines obtained for all the scan lines may be set as Xr3, and (Xr3-Xra) may be set as the changed horizontal integration threshold.

Embodiment 10 In the same manner as in Embodiment 8 , in the line generation processing, when there is a vertical ruled line rectangle overlapping with the character rectangle of interest, the horizontal integration threshold value is changed. Are different. Except for this, it is the same as the eighth embodiment .

FIG. 19 is an explanatory diagram of a method of determining a change value of the horizontal integration threshold. In FIGS. 19 (a) and (b),
θ is the skew angle of the document image measured in advance with reference to the Y axis (vertical direction). The measurement of the skew angle may be obtained by a well-known method such as Hough transform, or by a method of calculating from the difference between the maximum height of a character rectangle belonging to the same line and the line height. Here, the following cases are classified according to the inclination direction of the skew with respect to the Y axis.

(A) is the case where θ ≧ 0, and is obtained from the diagonal vertex coordinates of the rectangle A and the diagonal vertex coordinates of the vertical ruled line rectangle 386.
The change value Th1 'of the horizontal integration threshold is calculated by the following equation: Th1' = {Yr2- (Ya1 + Ya2) / 2} tanθ + Xr1-Xa2.

(B) is a case where θ <0, and is obtained from the diagonal vertex coordinates of the rectangle A and the diagonal vertex coordinates of the vertical ruled line rectangle 387.
The change value Th1 'of the horizontal integration threshold is calculated by the following equation: Th1' = {(Ya1 + Ya2) / 2-Yr1} tanθ + Xr1-Xa2.

Embodiment 11 The above-described embodiment 8 is performed up to the line generation processing by integrating the character rectangles.
10 to 10 , but a character area is generated by the character area generation unit 310 by integrating the lines. FIG. 20 shows a flow of the character area generation processing by line integration.

The first step 400 in FIG. 20 is processing at the stage of row generation. In this step, the target rectangle A
When there is a vertical ruled line rectangle on the right side within a certain distance range (that is, when there is a vertical ruled line that caused the horizontal integration threshold to be changed at the time of row generation in the eighth to tenth embodiments ) , A mark R indicating that a vertical ruled line exists on the right side is added to the generated line (step 4).
00). For example, a mark R is added to the row 381 in FIG.

Step 402 and subsequent steps are processing for generating a character area by line integration. First, a new row A is selected (step 402). A row whose horizontal distance and vertical distance from the noted row A are equal to or less than a predetermined value is searched, and if found, it is checked whether the row is a row with a mark R (step 404). If the line is not a line with the mark R, the lines are integrated (step 406), and the search for the next line is performed.

If the searched row is a row with a mark R, it is determined whether the horizontal end coordinate (Xr2) of this row and the horizontal end coordinate (Xa2) of row A satisfy the relationship of Xa2> Xr2. (Step 412). If this relationship is not satisfied, row integration is performed (step 406).
However, when the relationship is satisfied, the process proceeds to the search for the next line without performing the line integration.

FIG. 21 shows an example in which row integration is not performed because the condition of step 412 is satisfied. In FIG. 21A, since the row 413 has the mark R,
It is not integrated with the line of interest A414. Row 41 like this
3, 414 can be considered as rows belonging to different columns separated by the vertical ruled line 416.

In FIG. 21B, the row with mark R 417 and the row of interest A 418 are not integrated. Such rows 417 and 418 may not belong to different columns, but judging from the positional relationship with the vertical ruled line 419, it is natural that they are not integrated.

Embodiment 12 In the same way as in Embodiment 11 , processing up to the generation of a character area by line integration is performed, but a mark R is also added to a character area where a mark R is added to the integrated line. Then, the character area generated by the line integration is integrated by the character area correction unit 311 by the same processing as the line integration, whereby
In the row integration, the areas that could not be integrated are re-integrated.
In this reintegration, the character area to be integrated is marked R
Checked whether it is attached, the examples in the case of marked R 1
Similarly to the case of the line integration of No. 1 , when the horizontal end coordinate Xr2 of the region with the mark R and the horizontal end coordinate Xa2 of the focused character region are in a relationship of Xa2> Xr2, they are not integrated ( See FIG. 21).

FIG. 22 is a block diagram for explaining Embodiments 13 to 16 to be described later. Reference numeral 501 denotes a document image memory for storing input document image data; 506, a rectangle extracting unit for extracting a circumscribed rectangle of a black connected component of the input document image; 502, a rectangular memory for storing information on the extracted rectangle; is there. The rectangle extraction process may be performed on an image obtained by compressing the input document image. A rectangle classification unit 507 classifies rectangles into character rectangles and other figure / table rectangles. The processing of this rectangle classification is performed according to the eighth to thirteenth embodiments.
12 or the same as in the first to seventh embodiments.

Reference numeral 808 denotes a horizontal ruled line identification unit for identifying a horizontal ruled line (horizontal separator). This horizontal ruled line identification
For example, the following process is performed. Among the rectangles classified as figure / table rectangles (rectangles other than character rectangles) by the processing of the rectangle classification unit 507, the height (vertical size) is smaller than a predetermined threshold and the width (vertical direction) is smaller. Is selected as a candidate for a horizontal ruled line. Next, by scanning the document image in the range of the horizontal ruled line candidate rectangle in the horizontal direction, only black runs longer than a predetermined threshold are extracted, and a circumscribed rectangle of a connected component of only the long black runs is extracted. Then, between the height H1 and width W1 of the circumscribed rectangle of the long black run connected component and the height H and width W of the original rectangle, H1 / H> threshold (for example, 0.8) and W1 / If W> threshold (for example, 0.8), it is determined that the circumscribed rectangle of the long black run connected component is a horizontal ruled line. Note that the threshold value for determining the horizontal ruled line rectangle and the threshold value for the length of the black run may be adaptively determined in the same manner as the threshold value of the rectangle classification.

Reference numeral 509 denotes a line extraction unit that performs integration determination on the horizontal and vertical distances of the character rectangles based on the integration threshold and integrates the character rectangles that satisfy the integration conditions to generate a line. However, the rectangle of the horizontal ruled line extracted by the horizontal ruled line identification unit 508 is also extracted as an imaginary line. The integration threshold is given in advance or is determined adaptively based on a histogram of the height of the rectangle. A row memory 503 stores information on the extracted rows.

Reference numeral 510 denotes a character area generation unit that integrates the generated lines to generate a character area. Reference numeral 505 denotes an area memory for storing information on the generated character area. The process of line integration is performed by comparing the distance between lines to be integrated with an integration threshold to determine whether or not integration is possible, and integrating possible lines. In this integration determination, horizontal ruled line identification is performed. The integration threshold is changed according to the presence of the horizontal ruled line extracted by the unit 508. An integration threshold memory 504 stores an integration threshold for row integration. 5
Reference numeral 11 denotes a character region correction unit that corrects a character region by reintegrating a region that has not been completely integrated into a character region generated by the line integration of the character region generation unit 510. 5
Reference numeral 12 denotes a control unit that controls the above-described units.

Hereinafter, Examples 13 to 16 will be described individually. However, it is assumed that the lines on the document are horizontal.

Embodiment 13 A rectangle extraction unit 506 extracts a rectangle, a rectangle classification unit 507 classifies a rectangle, a horizontal ruled line identification unit 508 extracts a vertical ruled line, and a line extraction unit 509 extracts a line. Character area generation is performed.

FIG. 23 shows a schematic flow of the character area generating process. The basic processing for generating the character area is performed by determining the horizontal distance (p1) and the vertical distance (p2) between the line of interest and another line, the standard character size, and the like in advance.
3 is compared with the horizontal integration threshold value (Th1) and the vertical integration threshold value (Th2) stored in No. 3 and rectangles closer than the integration threshold value are integrated. As shown in FIG. An operation of changing the vertical integration threshold is performed in relation to the row.

That is, a new row A is selected (step 520), and the vertical integration threshold is set to Th2 (step 52).
2). Below the noted row A, the vertical distance is within the vertical integration threshold Th2 and the horizontal distance is the horizontal integration threshold Th1.
A line (including a horizontal ruled line rectangle as an imaginary line) is searched within the range of, and it is determined whether or not the found line is a horizontal ruled line (step 524). If the line is the original line, it is integrated with the line of interest A (step 526), and the search for the next line is performed.

If the searched line is a horizontal ruled line (an imaginary line), the line is naturally not integrated with the horizontal ruled line. Then, it is checked whether or not the line of interest A and the horizontal ruled line overlap in the horizontal direction (step 530). If there is an overlap, the distance Th2 from the line of interest A to the center line of the horizontal ruled line is determined. Is reset to the vertical integration threshold (step 532), and a search for the next rectangle is performed.

When the processing is repeated for the row A until the next row is not found within the vertical distance range equal to or less than the vertical integration threshold (Th1 or Th1 ') (step 526), the processing starts from step 520 for another target row. I do.

FIG. 24 shows a specific example when a horizontal ruled line exists.
This will be described below. In FIG. 24A, when the horizontal ruled line 542 is encountered during the line integration for the row A, this vertical ruled line 542 overlaps with the row A, so that the vertical integration threshold is changed from Th2 to Th2 ′. If the vertical integration threshold is Th2, row B 543, which was to be searched and integrated next, is no longer integrated with row A. Thus, row A
Character area 544 including a horizontal ruled line (horizontal separator)
The row B 543 separated by 542 is correctly separated.

In FIG. 24B, a horizontal ruled line 546 is searched for row A. However, this horizontal ruled line 5
Since the line 46 does not overlap the line of interest A 540 in the horizontal direction, the vertical integration threshold value remains at Th2. Therefore, the next line B543 is searched and integrated with line A540, and the character area 54 including line A540 and line B543 is searched.
8 is generated. That is, since the horizontal ruled line 546 is not a horizontal separator at least between the row A 540 and the row B 543, the existence of the horizontal ruled line 546 is ignored and the line integration is performed.

Embodiment 14 In the same manner as in Embodiment 13 , in the process of character area generation (line integration), if there is a horizontal ruled line overlapping the line of interest, the vertical integration threshold is changed. The method of determining the value is different. Except for this, it is the same as the thirteenth embodiment .

FIG. 25 is an explanatory diagram of a method for determining a change value of the vertical integration threshold. The horizontal range (Xa1 to Xa2) of the line A550 of interest and the rectangle 551 of the horizontal ruled line
The image is vertically scanned in a scan range 552 determined by the vertical range (Yr1 to Yr2) of
Existence position of a horizontal ruled line on a certain scan line (Yr3)
And the distance (Yr3-Ya2) is set as the changed vertical integration threshold. Alternatively, assuming that the average value of the horizontal ruled line positions obtained for all the scan lines is Yr3, (Yr3-Y
ra) may be set as the changed vertical integration threshold.

Fifteenth Embodiment As in the thirteenth embodiment , the vertical integration threshold is changed when there is a horizontal ruled line overlapping the line A of interest in the character area formation processing. Are different. Except for this, it is the same as the thirteenth embodiment .

FIG. 26 is an explanatory diagram of a method of determining a changed value of the vertical integration threshold. In FIGS. 26A and 26B,
θ is the skew angle of the document image measured in advance with reference to the X axis (horizontal direction). The measurement of the skew angle is as described above.
What is necessary is just to carry out by the method as described in the tenth embodiment .
Here, the following cases are classified according to the inclination direction of the skew with respect to the X axis. (A) is a case where θ ≧ 0, and is obtained from the diagonal vertex coordinates of the rectangle of interest A553 and the diagonal vertex coordinates of the rectangle 554 of the horizontal ruled line.
The change value Th2 'of the vertical integration threshold is calculated by the following equation: Th2' = {(Xa1 + Xa2) / 2-Xr1} tanθ + Yr1-Ya2.

(B) shows the case where θ <0, and the rectangle A55
From the diagonal vertex coordinates of No. 3 and the diagonal vertex coordinates of the horizontal ruled line rectangle 555, the change value Th2 'of the vertical integration threshold is calculated by the following equation: Th2' = {Xr2- (Xa1 + Xa2) / 2} tanθ + Yr1-Ya2.

Embodiment 16 The processing up to the generation of a character area by line integration is the same as that of the embodiment 13
15 to 15 , but the character area correcting unit 511 performs a reintegration process for character areas that could not be integrated. FIG. 27 shows a schematic flow of this processing.

The first step 560 in FIG. 27 is a process at the stage of row integration. In this step, if a horizontal ruled line exists below the line of interest A within a certain distance range, “the horizontal ruled line exists below” in the generated character area.
Is added (step 56).
2). For example, in the character area 544 of FIG.
Is attached.

Step 562 and subsequent steps correspond to the character area reintegration processing. First, a new character area A is selected (step 562). A character area whose horizontal distance and vertical distance from the target character area A are equal to or smaller than a predetermined value is searched, and if found, it is checked whether the character area is an area with a mark R (step 563). If the line is not marked R,
Integration is performed (step 564), and a search for the next character area is performed.

If the searched character area is an area with a mark R, the end point coordinates (Yr
It is determined whether 2) and the end point coordinates (Ya2) of the character area A in the vertical direction satisfy the relationship of Ya2> Yr2 (step 567). When this relationship is not satisfied, integration is performed (step 564). When the relationship is satisfied, integration is not performed and the process proceeds to the search for the next character area.

FIG. 28 shows an example in which integration is not performed because the condition of step 567 is satisfied. Character area 569
Is marked with a mark R and is not integrated with the noted character area A568. Such character areas can be considered as separate areas separated by horizontal ruled lines.

FIG. 32 is a view for explaining Example 17 to be described later. FIG. 32A shows an example of an input document.
In this document, 660 is a table area surrounded by vertical and horizontal ruled lines, 661 is a graph area, and shaded parts are character areas (areas of one or more characters or lines).

Conventionally, when a character area is extracted from this document, as shown in FIG. 32B, the graph area 661 and the table area 660 and the character area adjacent thereto are likely to be mistakenly integrated.

According to the seventeenth embodiment , by preventing such erroneous integration of character regions, it becomes possible to divide the document shown in FIG. 32A into regions as shown in FIG. 32C. . That is, according to Example 17 described later, FIG.
As shown in (c), the lowermost horizontal ruled line 663 of the graph area 661 and the uppermost horizontal ruled line 66 of the table area 660 are displayed.
By recognizing 4 as an imaginary horizontal separator for separating a character area (column), the upper and lower character areas are separated from the boundary.

Further, according to the eighteenth embodiment , by recognizing a vertical imaginary separator, erroneous integration between a vertically adjacent figure / table area and a character area is prevented.

FIG. 29 is a block diagram for explaining Embodiments 17 and 18 to be described later. Referring to FIG.
Is a document image memory for storing input document image data, 606 is a rectangle extracting unit that extracts a circumscribed rectangle of a black connected component of the input document image, and 602 is a rectangle memory that stores information on the extracted rectangle. . A rectangle classification unit 607 classifies the extracted rectangle into a character rectangle and other large rectangles such as graphs and tables. Information on this classification result is also stored in the rectangular memory 602.

Reference numeral 608 denotes a ruled line extraction unit for extracting a horizontal ruled line or a vertical ruled line from a figure / table rectangle. A ruled line memory 603 stores information on the extracted ruled lines. Reference numeral 609 denotes an imaginary separator testing unit that determines whether the extracted horizontal or vertical ruled line is valid as a horizontal or vertical imaginary separator, and 604 denotes an imaginary separator memory that stores information on the imaginary separator. Reference numeral 610 denotes a character area generation unit that generates a character area, and uses a fictitious separator as an actual separator when integrating areas for generating a character area. An area memory 605 stores information of the generated character area. A control unit 611 controls the above units.

Hereinafter, Examples 17 and 18 will be described individually.

Embodiment 17 FIG. 30 shows a schematic flow of a process for extracting an imaginary separator. From the rectangles extracted by the rectangle extraction unit 606, the width (the size in the horizontal direction) W and the height H are W>
A rectangle (1) of LARGEHTH and H> LARGEVTH is selected (steps 620 and 621). This is the processing of the rectangle classification unit 607. This rectangle (1) is shown in FIG.
The table area 660 and the graph area 661 of the document shown in FIG.
It corresponds to a large area other than the character like.
Note that the threshold values LARGEHTH and LARGEVTH are
It is a value given in advance or a value adaptively determined from a histogram of the height of the rectangle.

Next, for the rectangle (1), the ruled line extracting unit 60
In step 8, a horizontal ruled line is extracted (step 622). More specifically, the range of the image occupied by the rectangle (1) is scanned in the horizontal direction to detect a black run, a black run having a length equal to or longer than the threshold value RUNHTH is extracted, and a circumscribed component is connected to only the long black run connected component. The extracted rectangle (2) is extracted as a horizontal ruled line.
At this time, the width W1 and the height H1 of the rectangle (2) are also extracted.
Information on the extracted rectangle (2) is stored in the ruled line rectangle memory 603. In the rectangle (1) including a plurality of horizontal ruled lines, such as the table area 660, a plurality of rectangles (2) are extracted. The threshold value RUNHTH is a value given in advance or a value adaptively determined from a histogram of the height of the rectangle.

Next, in the fictitious separator verification section 609, it is verified whether the horizontal ruled line rectangle (2) extracted from the rectangle (1) is valid as a fictitious horizontal separator (steps 623 and 624). The rectangle (2) that satisfies all the following conditions a to c is determined to be a fictitious horizontal separator, and the information is stored in the fictitious separator memory 604.

A) W1 / W> threshold value (for example, 0.8) (W is the width of the original rectangle (1)) b) H1 <threshold value RLHeighthTH (RLHeightTH is a predetermined value or is a certain value.) C) The rectangle (2) is upper or lower.

The condition c) will be described with reference to FIG. FIG. 31A shows an example of a table area as a rectangle (1). (B) is an explanatory diagram of upper and lower, and 630 to 632 show the horizontal ruled line rectangle (2). The uppermost rectangle (2) in the range RangeUTH from the upper side of the rectangle (1) is upper. That is, in this example, the rectangle 630 is the upper. The lowermost rectangle (2) in the range RangeLTH from the lower side of the rectangle (1) is the lower. In this example, the rectangle 632 is lower.

In the character area generation unit 610, the imaginary horizontal separator extracted by the above processing is
The same as the horizontal ruled lines (horizontal separators) in Embodiments 13 to 16 , the same as in Embodiments 13 to 16,
A character area is generated by integrating character rectangles by the same processing as that described above, and information on the generated character area is stored in the area memory 605.
To be stored.

Embodiment 18 In this embodiment, a vertical imaginary separator is extracted.
That is, the overall flow of the process of extracting the imaginary separator is shown in FIG. 23, but the contents of steps 622 to 625 are changed according to the difference in the direction of the separator.

That is, in step 622, an image within the range of a large rectangle (1) other than a character is scanned in the vertical direction to detect a black run, and is converted into a connected component of only the black run having a length equal to or longer than the threshold value RUNTH. The circumscribed rectangle (2) is extracted. That is, the rectangle of the vertical ruled line is extracted.

In steps 623 to 625, the rectangle (2) as a vertical ruled line extracted from the rectangle (1) is
Test the validity as a fictitious vertical separator. Next d
The rectangle (2) that satisfies all of the conditions (f) to (f) is determined as the imaginary vertical separator.

D) H1 / H> threshold (for example, 0.8) (where H is the height of the original rectangle (1)) e) W1 <threshold RLWidthTH (RLWidth is a predetermined value, or F) The rectangle (2) is left or right.

The condition f) will be described with reference to FIG. FIG. 31C is an explanatory diagram of left and right, and 634 to 636 show a rectangle (2) of a vertical ruled line. The leftmost rectangle (2) in the range RangeLTH from the left side of the rectangle (1) is left. In this example, the rectangle 634 is left. Also, the rightmost rectangle (2) in a range RangeRTH from the right side of the rectangle (1)
Is right. In this example, the rectangle 635 is rig
ht.

In the character area generation unit 610, the imaginary vertical separator extracted by the above processing is
Vertical ruled lines (vertical separator) in Examples 8 to 12
In the same manner as in Examples 8 to 12 , character rectangles are integrated to generate a character area by the same processing as in the eighth to twelfth embodiments , and information on the generated character area is stored in the area memory 605.

[0123]

According to the first to fourth aspects of the present invention, it is possible to accurately classify character rectangles and other rectangles (chart rectangles) in document images of various character sizes. Therefore, the extraction accuracy of a character area, which is a set of character rectangles, is improved. According to the fourth aspect of the present invention, even for a document image read from a book or a thick original, the influence of a black area generated in a bent portion or a peripheral part is eliminated, and a character rectangle, a line, or a character area can be accurately determined. Can be extracted.

According to the fifth to ninth aspects of the present invention, erroneous integration of character areas (columns) separated by a vertical separator or a horizontal separator can be prevented, and each character area can be correctly separated. Further, the processing contents are simple, the processing can be speeded up easily, and the amount of memory required for the processing can be reduced.

According to the tenth or eleventh aspect of the present invention, it is possible to prevent erroneous integration between a table or graph area and a character area approaching the area, and greatly improve the accuracy of area division.

[Brief description of the drawings]

FIG. 1 is a block diagram for explaining Examples 1 to 6.

FIG. 2 is a diagram for explaining an outline of processing.

FIG. 3 is a processing flowchart from rectangle extraction to standard character size determination in the first embodiment.

FIG. 4 is a diagram showing an example of a histogram of rectangular heights;
You.

FIG. 5 is an explanatory diagram of character rectangle integration in a second embodiment.

FIG. 6 is a diagram showing lines extracted by integrating character rectangles.

FIG. 7 is an explanatory diagram of line extraction by character rectangle integration in a third embodiment.

FIG. 8 is an explanatory diagram of character region extraction by line integration according to a fifth embodiment.

FIG. 9 is an explanatory diagram of character region extraction by line integration according to a fifth embodiment.

FIG. 10 is an explanatory diagram of integration of a large character area according to a sixth embodiment.

FIG. 11 is an explanatory diagram of integration of small character areas according to a sixth embodiment.

FIG. 12 is a block diagram for explaining a seventh embodiment ;

FIG. 13 is a flowchart of noise rectangle removal processing according to a seventh embodiment.

FIG. 14 is an explanatory diagram of noise rectangle removal in a seventh embodiment .

FIG. 15 is a block diagram for explaining Examples 8 to 12 ;

FIG. 16 is a flowchart of a row generation process according to the eighth embodiment .

FIG. 17 shows a relationship between row integration and vertical ruled lines in the eighth embodiment .
FIG.

FIG. 18 is an explanatory diagram of a method of obtaining a distance to a vertical ruled line according to the ninth embodiment .

FIG. 19 is a diagram illustrating a distance to a vertical ruled line according to the tenth embodiment .
FIG.

FIG. 20 is a flowchart of a row integration process according to the eleventh embodiment .

FIG. 21 is an explanatory diagram of row integration in the eleventh embodiment .

FIG. 22 is a block diagram for explaining Embodiments 13 to 16 ;

FIG. 23 is a flowchart of a row integration process according to the thirteenth embodiment .

FIG. 24 is a diagram showing a relation between row integration and horizontal ruled lines in the thirteenth embodiment .
It is explanatory drawing of a person in charge.

FIG. 25 is an explanatory diagram of a method of obtaining a distance to a horizontal ruled line according to the fourteenth embodiment .

FIG. 26 is a diagram illustrating a distance to a horizontal ruled line according to the fifteenth embodiment;
FIG.

FIG. 27 is a flowchart of a character area reintegration process according to the sixteenth embodiment .

FIG. 28 is an explanatory diagram of character area reintegration in the sixteenth embodiment .

FIG. 29 is a block diagram for explaining Embodiments 17 and 18 ;

FIG. 30 is a flowchart of an imaginary horizontal separator extraction process according to the seventeenth embodiment .

FIG. 31 is an imaginary separator according to the seventeenth and eighteenth embodiments .
It is explanatory drawing of lator extraction.

FIG. 32 is an explanatory diagram of prevention of erroneous integration of character areas according to the seventeenth embodiment .

[Explanation of symbols]

 1,301,501,601 Document image memory 2,302,502,602 Rectangular memory 3 Histogram memory 4 Standard character size holding memory 5,306,506,606 Rectangular extracting unit 6 Histogram generating unit 7 Standard character size determining unit 8, 307, 507, 607 Rectangle classification unit 9, 309, 509 Line extraction unit 10, 310, 510 Character region generation unit 11, 312, 512, 611 Control unit 12, 304, 503 Line memory 13, 305, 505, 605 Region memory 201 Noise rectangle removal unit 303,504 Integrated threshold memory 308 Vertical ruled line identification unit 311,211 Character area correction unit 508 Horizontal ruled line identification unit 603 Ruled line memory 604 Imaginary separator memory 608 Ruled line extraction unit 609 Imaginary separator test unit 610 Character region generation unit

Claims (11)

(57) [Claims] 1. A rectangle circumscribing a black connected component is extracted from a document image, a standard character size is determined from a histogram of the height of the extracted rectangle, and the extracted rectangle is determined by the size relation between the size and the standard character size. Is classified into character rectangles and other rectangles based on the character rectangle, and it is determined whether or not to combine the character rectangles based on the distance between the character rectangles and the standard character size. A region dividing method characterized by extracting a circumscribed rectangle as a line of a character string. 2. The area dividing method according to claim 1, wherein
An area classification method characterized by determining whether or not to combine character rectangles based on the overlap amount between character rectangles and the height of character rectangles, in addition to the distance between character rectangles and the standard character size. 3. A rectangle circumscribing a black connected component is extracted from a document image, a standard character size is determined from a histogram of the heights of the extracted rectangles, and the extracted rectangle is determined by the size relationship between the size and the standard character size. Are classified into character rectangles and other rectangles based on the character strings, the character rectangles are integrated, and a circumscribed rectangle of the integrated character rectangle group is extracted as a character string line. An area dividing method characterized by determining whether to integrate based on a distance between lines and the standard character size, and extracting the integrated character string line as a character area. 4. The method of claim 1, 2 or 3 the area dividing method according, in contact with the boundary of the processing region of the document image
Area dividing method characterized by removing rectangular except that character as noise. 5. A region dividing method for extracting a rectangle circumscribing a black connected component corresponding to a character from a document image and generating a line by integrating the rectangle of the extracted character, extracting a vertical ruled line from the document image, When merging rectangles from a given rectangle, a vertical ruled line whose horizontal distance to the rectangle of interest is smaller than the threshold value of the distance between rectangles, which is a horizontal integration condition, and which overlaps the rectangle of interest vertically Wherein the threshold value of the distance between rectangles, which is a horizontal integration condition, is changed to a value corresponding to the horizontal distance between the rectangle of interest and the vertical ruled line for the region of interest. . 6. An area division for extracting a rectangle circumscribing a black connected component corresponding to a character from a document image, generating a line by integrating the extracted character rectangles, and generating a character area by integrating the generated lines. In the method, a vertical ruled line is extracted from a document image, and when a rectangle is integrated from a certain rectangle of interest, a horizontal distance to the rectangle of interest is smaller than a threshold value of a distance between rectangles, which is a horizontal integration condition, and When a vertical ruled line having a vertical overlap with the target rectangle is encountered, the threshold value of the distance between rectangles, which is a horizontal integration condition, for the target region is set to correspond to the horizontal distance between the target rectangle and the vertical ruled line. When merging lines from a certain line of interest, the line to be merged changes the threshold of the distance between rectangles, which is the horizontal merging condition, by encountering a vertical ruled line at the time of generation. In some cases, when the horizontal end point coordinates of the line of interest are larger than the horizontal end point coordinates of the line to be integrated, the integration of the lines to be integrated is prohibited. 7. A method according to claim 6 , wherein:
The character area generated by the line integration is re-integrated from the character area of interest, and at this time, a horizontal line in a line in the character area to be integrated is encountered due to encounter with the vertical ruled line at the time of generation. If a line in which the threshold value of the distance between rectangles as a direction integration condition is changed is included, and the horizontal end point coordinate of the target character area is larger than the horizontal end point coordinate of the character area to be integrated, A region dividing method characterized by prohibiting the integration of character regions to be integrated. 8. An area division for extracting a rectangle circumscribing a black connected component corresponding to a character from a document image, generating a line by integrating the extracted character rectangle, and generating a character area by integrating the generated lines. In the method, a horizontal ruled line is extracted from a document image, and when lines are integrated from a certain line of interest, a vertical distance from the line of interest is smaller than a threshold of a line-to-line distance that is a vertical integration condition, and When a horizontal ruled line having a horizontal overlap with the line of interest is encountered, the threshold value of the distance between lines, which is a vertical integration condition, is set to a value corresponding to the vertical distance between the line of interest and the horizontal ruled line for the line of interest. A region dividing method characterized by changing. 9. The area dividing method according to claim 8 , wherein
The character area generated by the line integration is re-integrated from the character area of interest, and at this time, a line in the character area to be integrated is placed vertically by encountering the horizontal ruled line when it is generated. If a line in which the threshold value of the line-to-line distance as a direction integration condition is changed is included, and if the vertical end coordinate of the target character region is larger than the vertical end coordinate of the character region to be integrated, the integration is performed. A region dividing method characterized in that integration of character regions to be attempted is prohibited. 10. A method according to claim 8 , wherein a rectangle circumscribing a large black connected component other than the character is extracted from the document image, and the rectangle other than the extracted character is scanned in the horizontal direction to obtain a long rectangle equal to or larger than the threshold value. A rectangle H circumscribing the black connected component consisting of only the black run is extracted, and among the rectangles H extracted from the rectangles other than the respective characters, the rectangle H extracted from the upper side or the lower side of the rectangle other than the character within a predetermined distance range. Or, the one that satisfies the shape condition as the horizontal ruled line and is extracted as a fictitious horizontal separator is extracted.When the character area is generated by integrating the character rectangles, the fictitious horizontal separator is moved up and down the character area. An area dividing method, wherein the area is treated as the horizontal ruled line to be separated. 11. The area dividing method according to claim 6 , wherein a rectangle circumscribing a large black connected component other than a character is extracted from a document image, and the extracted rectangle other than the character is scanned in a vertical direction to obtain a longer rectangle than a threshold. A rectangle V circumscribing the black connected component consisting of only the black run is extracted, and among the rectangles V extracted from the rectangles other than the respective characters, the leftmost within a predetermined distance from the left side or the right side of the rectangle other than the character is extracted. Or, the one that satisfies the shape condition as the vertical ruled line and is extracted as the imaginary vertical separator is extracted as the imaginary vertical separator. An area dividing method, wherein the area is treated as the vertical ruled line to be divided.